Speech recognition systems typically aim to convert a voice input from a user into a natural language transcription. Speech recognition systems may use an acoustic model in the process of determining the words that a user has spoken. For example, an acoustic model may include statistical information that can be used to estimate which speech sounds occur in audio.